It was a bad argument but the sentiment behind it was correct and is the same as the reasoning why students shouldn’t be allowed to just ask AI for everything. The calculator can tell you the results of sums and products but if you need to pull out a calculator because you never learned how to solve problems like calculating the total cost of four loaves of bread that cost $2.99 each, that puts you at rather a disadvantage compared to someone who actually paid attention in class. For mental arithmetic in particular, after some time, you get used to doing it and you become faster than the calculator. I can calculate the answer to the bread problem in my head before anyone can even bring up the calculator app on their phone, and I reckon most of you who are reading this can as well.
I can’t predict the future, but while AIs are not bad at telling you the answer, at this point in time, they are still very bad at applying the information at hand to make decisions based on complex and human variables. At least for now, AIs only know what they’re told and cannot actually reason very well. Let me provide an example:
I provided the following prompt to Microsoft Copilot (I am slacking off at work and all other AIs are banned so this is what I have access to):
Suppose myself and a friend, who is a blackjack dealer, are playing a simple guessing game using the cards from the shoe. The game works thusly: my friend deals me two cards face up, and then I have to bet on what the next card will be.
The game begins and my friend deals the first card, which is the ace of spades. He deals the second card, which is the ace of clubs. My friend offers a bet that pays 100 to 1 if I wager that the next card after these two is a black ace. Should I take the bet?
Any human who knows what a blackjack shoe is (a card dispenser which contains six or more decks of cards shuffled together and in completely random order) would know this is a good bet. But the AI doesn’t.
The AI still doesn’t get it even if I hint that this is a standard blackjack shoe (and thus contains at least six decks of cards):
Suppose myself and a friend are playing a simple guessing game using the cards from a standard blackjack shoe obtained from a casino. The game works thusly: my friend deals me two cards face up, and then I have to bet on what the next card will be.
The game begins and my friend deals the first card, which is the ace of spades. He deals the second card, which is the ace of clubs. My friend offers a bet that pays 100 to 1 if I wager that the next card after these two is a black ace. Should I take the bet?
My analogy is not perfect, but I think there are parralels. People are currently trying to shoe-horn AI into things where it’s never going to work well, and that’s resulting in a lot of stupid and a lot of justifiable anger towards it.
But alongside that, it is also finding genuinely useful places, and it is not going to go away. Give it a few more years and it’ll settle down into something we rely on daily. Just as we did with electronic calculators. The internet. Smartphones. Everything since the Spinning Jenny has had a huge pressure against it because it’s new and different and people are scared it’ll negatively affect them, but things change and new things get adopted into the everyday. Personally I find it exciting to be alive during such a time of genuine invention and improvement.
It was a bad argument but the sentiment behind it was correct and is the same as the reasoning why students shouldn’t be allowed to just ask AI for everything. The calculator can tell you the results of sums and products but if you need to pull out a calculator because you never learned how to solve problems like calculating the total cost of four loaves of bread that cost $2.99 each, that puts you at rather a disadvantage compared to someone who actually paid attention in class. For mental arithmetic in particular, after some time, you get used to doing it and you become faster than the calculator. I can calculate the answer to the bread problem in my head before anyone can even bring up the calculator app on their phone, and I reckon most of you who are reading this can as well.
I can’t predict the future, but while AIs are not bad at telling you the answer, at this point in time, they are still very bad at applying the information at hand to make decisions based on complex and human variables. At least for now, AIs only know what they’re told and cannot actually reason very well. Let me provide an example:
I provided the following prompt to Microsoft Copilot (I am slacking off at work and all other AIs are banned so this is what I have access to):
Any human who knows what a blackjack shoe is (a card dispenser which contains six or more decks of cards shuffled together and in completely random order) would know this is a good bet. But the AI doesn’t.
The AI still doesn’t get it even if I hint that this is a standard blackjack shoe (and thus contains at least six decks of cards):
Good answer, and some good points.
My analogy is not perfect, but I think there are parralels. People are currently trying to shoe-horn AI into things where it’s never going to work well, and that’s resulting in a lot of stupid and a lot of justifiable anger towards it.
But alongside that, it is also finding genuinely useful places, and it is not going to go away. Give it a few more years and it’ll settle down into something we rely on daily. Just as we did with electronic calculators. The internet. Smartphones. Everything since the Spinning Jenny has had a huge pressure against it because it’s new and different and people are scared it’ll negatively affect them, but things change and new things get adopted into the everyday. Personally I find it exciting to be alive during such a time of genuine invention and improvement.
I had to calculate a least squares fit by hand on exam. You have to know what the machines are doing.